A study of Spoken Word Recognition using Unsupervised Learning with reference to Assamese Language

The aim of this paper is to measure the performance of Auto Associative Neural Network (AANN) in terms of recognition rate of spoken words with reference to Assamese language. One AANN models is built in the present study for spoken word recognition using continuous speech database. The recognition is measured by taking Mel Frequency Cepstral Coefficients (MFCC) as the feature vector along with its first and second derivatives. In this study, one database is created by using continuous speech. In the continuous database, there are 240 continuous speech files each around two minutes of duration to train and test the AANN model. AANN is trained using the continuous speech database and the recognition of certain words present within the continuous speech signal is measured. In this study, fifteen native Assamese speakers are used which involves eight male speakers and seven female speakers. It is observed that the performance of AANN is 85.08% in terms of accuracy and the performance of AANN is 93.61% in terms of sensitivity.